📦 Parsera

Lightweight Python library for scraping websites with LLMs. You can test it on Parsera website.

Why Parsera?

Because it's simple and lightweight, with minimal token use which boosts speed and reduces expenses.

Installation

pip install parsera
playwright install

Basic usage

If you want to use OpenAI, remember to set up OPENAI_API_KEY env variable. You can do this from python with:

import os

os.environ["OPENAI_API_KEY"] = "YOUR_OPENAI_API_KEY_HERE"

Next, you can run a basic version that uses gpt-4o-mini

from parsera import Parsera

url = "https://news.ycombinator.com/"
elements = {
    "Title": "News title",
    "Points": "Number of points",
    "Comments": "Number of comments",
}

scraper = Parsera()
result = scraper.run(url=url, elements=elements)

result variable will contain a json with a list of records:

[
   {
      "Title":"Hacking the largest airline and hotel rewards platform (2023)",
      "Points":"104",
      "Comments":"24"
   },
    ...
]

There is also arun async method available:

result = await scrapper.arun(url=url, elements=elements)

Using proxy

You can use serve the traffic via proxy server when calling run method:

proxy_settings = {
    "server": "https://1.2.3.4:5678",
    "username": <PROXY_USERNAME>,
    "password": <PROXY_PASSWORD>,
}
result = scrapper.run(url=url, elements=elements, proxy_settings=proxy_settings)

Run with custom model

You can instantiate Parsera with any chat model supported by LangChain, for example, to run the model from Azure:

import os
from langchain_openai import AzureChatOpenAI

llm = AzureChatOpenAI(
    azure_endpoint=os.getenv("AZURE_GPT_BASE_URL"),
    openai_api_version="2023-05-15",
    deployment_name=os.getenv("AZURE_GPT_DEPLOYMENT_NAME"),
    openai_api_key=os.getenv("AZURE_GPT_API_KEY"),
    openai_api_type="azure",
    temperature=0.0,
)

url = "https://news.ycombinator.com/"
elements = {
    "Title": "News title",
    "Points": "Number of points",
    "Comments": "Number of comments",
}
scrapper = Parsera(model=llm)
result = scrapper.run(url=url, elements=elements)

Run local model with HuggingFace `Trasformers`

Currently, we only support models that include a system token

You should install Transformers with either pytorch (recommended) or TensorFlow 2.0

Transformers Installation Guide

example:

from transformers import pipeline, AutoTokenizer, AutoModelForCausalLM
from parsera.engine.model import HuggingFaceModel
from parsera import Parsera

# Define the URL and elements to scrape
url = "https://news.ycombinator.com/"
elements = {
"Title": "News title",
"Points": "Number of points",
"Comments": "Number of comments",
}

# Initialize model with transformers pipeline
tokenizer = AutoTokenizer.from_pretrained("microsoft/Phi-3-mini-128k-instruct", trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained("microsoft/Phi-3-mini-128k-instruct", trust_remote_code=True)
pipe = pipeline("text-generation", model=model, tokenizer=tokenizer, max_new_tokens=5000)

# Initialize HuggingFaceModel
llm = HuggingFaceModel(pipeline=pipe)

# Scrapper with HuggingFace model
scrapper = Parsera(model=llm)
result = scrapper.run(url=url, elements=elements)

Using different extractor types

By default a tabular extractor is used, but you can also use the list or item extractors:

from parsera import Parsera

scraper = Parsera(extractor=Parsera.ExtractorType.LIST)
# or
scraper = Parsera(extractor=Parsera.ExtractorType.ITEM)

The tabular extractor is used to find rows of tabular data and has output of the form:

[
    {"name": "name1", "price": "100"},
    {"name": "name2", "price": "150"},
    {"name": "name3", "price": "300"},
]

The list extractor is used to find lists of different values and has output of the form:

{
    "name": ["name1", "name2", "name3"],
    "price": ["100", "150", "300"]
}

The item extractor is used to get singular items from a page like a title or price and has output of the form:

{
    "name": "name1",
    "price": "100"
}

Running with Jupyter Notebook:

Either place this code at the beginning of your notebook:

import nest_asyncio
nest_asyncio.apply()

Or instead of calling run method use async arun.

raznem/parsera